How to Automate a Greenhouse or Hydroponics System with OpenClaw on Raspberry Pi

A greenhouse concentrates growing variables into a small, interdependent system. Temperature affects humidity, which affects transpiration, which affects nutrient uptake, which changes EC and pH in the reservoir. Adjusting one variable ripples through the rest. A static threshold controller that treats each sensor independently misses these interactions entirely.

Hydroponics amplifies the problem. Plants growing without soil depend completely on the nutrient solution you provide. If the pH drifts outside the 5.5-6.5 range, nutrient lockout begins within hours. If EC climbs too high, roots burn. If it drops too low, growth stalls. Traditional automation handles this with fixed setpoints and dosing pumps, but it cannot account for why the pH shifted. Was it a hot day that increased evaporation and concentrated the solution? Did the plants enter a flowering stage and start consuming more phosphorus? A fixed controller does not ask these questions.

Hydroponics uses up to 90% less water than soil-based agriculture, according to research published in the National Institutes of Health's PMC database. Automated greenhouse systems can increase crop yield by 20-30% over manual management, based on studies reviewed by Springer Nature. But those gains depend on responsive automation, not just having sensors wired to relays.

OpenClaw changes the decision layer. Running on a Raspberry Pi alongside your sensors, the OpenClaw agent takes structured sensor input, considers the full environmental context, and makes a judgment call about what to adjust. Instead of "pH is 5.4, add base," the agent reasons: "pH dropped 0.3 in the last 4 hours, EC also rose by 0.2, and temperature spiked to 32C this afternoon. The solution is concentrating from evaporation, not from a dosing error. Top off with plain water before adjusting pH." That kind of reasoning previously required a grower checking the system in person.

A greenhouse or hydroponics monitoring setup needs four categories of sensors: nutrient solution quality, air environment, light levels, and water flow. Here is the full sensor stack, with wiring details for each.

pH sensor ($50-80): Atlas Scientific's EZO pH circuit is the standard for Pi-based hydroponics. It communicates over I2C (SDA to GPIO 2, SCL to GPIO 3) and returns calibrated pH readings accurate to 0.02 units. The probe sits in the nutrient reservoir. Cheaper pH modules exist, but they drift faster and need recalibration every few days. The Atlas board holds calibration for months.

EC/TDS sensor ($60-90): Electrical conductivity measures total dissolved nutrients. Atlas Scientific also makes an EZO conductivity circuit that shares the I2C bus with the pH board (each has a unique address). EC readings tell the agent whether the solution is too concentrated or too dilute. For most hydroponic crops, target EC ranges from 1.2-2.5 mS/cm depending on growth stage.

DHT22 or BME280 for air temperature and humidity ($5-10): The DHT22 connects to a single GPIO pin and returns temperature (accurate to 0.5C) and relative humidity. The BME280 connects over I2C and adds barometric pressure, which is useful for altitude-adjusted VPD (vapor pressure deficit) calculations. VPD matters because it directly controls transpiration rate, which drives nutrient uptake from the reservoir.

Water temperature sensor, DS18B20 ($3-5): Reservoir temperature affects dissolved oxygen levels and root health. The DS18B20 uses the 1-Wire protocol on GPIO 4 with a 4.7k pull-up resistor. Keep nutrient solution between 18-22C. Above 25C, dissolved oxygen drops and root pathogens like Pythium thrive.

Light sensor, BH1750 ($3-5): Connects over I2C and measures lux. Useful for tracking natural light levels so the agent can decide when to supplement with grow lights. Most fruiting crops need 12-16 hours of light above 300 umol/m2/s (roughly 20,000-40,000 lux depending on the light source).

Peristaltic dosing pumps ($10-20 each): Not sensors, but critical actuators. Each pump connects through a relay module to a GPIO pin. You need at minimum three pumps: pH-up solution, pH-down solution, and concentrated nutrient stock. The relay pattern is identical to the irrigation controller approach: signal pin to GPIO, pump power through the relay's normally-open contacts, separate power supply for the pumps.

Ventilation fans and grow lights: Both connect through relay modules. Exhaust fans control temperature and humidity. Grow lights supplement natural light during short days or overcast weather. Each relay channel maps to one GPIO pin.

All I2C devices share the same two-wire bus (GPIO 2 and GPIO 3) with unique addresses, so a single Raspberry Pi can read pH, EC, air conditions, and light level without running out of pins. The Adafruit Blinka compatibility layer gives you access to the same CircuitPython sensor libraries on a Pi that you would use on a microcontroller.

Hardware requirements: A Raspberry Pi 5 with 8GB RAM is recommended. The Pi 4 with 4GB works but leaves less headroom for concurrent sensor polling and LLM API calls. Use Raspberry Pi OS Lite (64-bit) for a headless setup. An M.2 SSD via the Pi 5's HAT+ connector is better than an SD card for reliability in an always-on greenhouse controller.

Install Adafruit Blinka and sensor libraries: Blinka provides the CircuitPython compatibility layer that makes sensor libraries work on a full Linux Pi. Install it alongside the specific libraries for your sensors:

pip3 install adafruit-blinka
pip3 install adafruit-circuitpython-bme280
pip3 install adafruit-circuitpython-bh1750

For Atlas Scientific pH and EC boards, use their own Python library, which communicates over I2C at the board's configured address. The default addresses are 0x63 for pH and 0x64 for conductivity, though these are configurable via the board's serial interface.

Sensor reading scripts: Create individual Python scripts that each return a single reading. A read_ph.py script queries the Atlas EZO board and prints the pH value. A read_ec.py returns conductivity in mS/cm. A read_environment.py reads the BME280 or DHT22 and returns temperature, humidity, and (if available) pressure. A read_reservoir_temp.py reads the DS18B20. Each script handles I2C communication and returns clean, structured output that the agent can parse.

Actuator control scripts: A dose_pump.py script accepts arguments for which pump (pH-up, pH-down, nutrient) and how long to run it (in seconds). Peristaltic pump flow rate is proportional to run time, so 5 seconds might deliver 2.5ml depending on the pump model. A set_fan.py and set_light.py script toggle the corresponding relays.

Install OpenClaw: Follow the official installation from the OpenClaw Raspberry Pi setup guide. OpenClaw's agent runtime uses roughly 200-400 MB of RAM. Configure it with your preferred LLM provider. Claude, GPT-4, Gemini, and local models through Ollama are all supported. For greenhouse automation, the decision loop runs every 5-15 minutes, so API costs stay low even with cloud-hosted models.

Agent configuration: The OpenClaw agent calls your sensor and actuator scripts as system commands. When it needs the current state of the greenhouse, it runs each sensor script and collects the readings. When it decides to dose nutrients or toggle a fan, it calls the corresponding actuator script. The agent handles the reasoning layer. The scripts handle hardware interaction. This separation means you can test and calibrate each sensor independently before connecting it to the agent's decision loop.

Store Your Greenhouse Data Where Any Agent Can Query It

Upload sensor logs, dosing records, and growth stage notes to a Fastio workspace. Intelligence Mode indexes everything for semantic search, so you can ask questions about your greenhouse's pH trends or nutrient consumption without parsing raw CSV files. generous storage, no credit card required. Built for openclaw raspberry greenhouse hydroponics automation agent workflows.

Start 14-Day Trial

The decision loop is where OpenClaw's reasoning separates this system from a PID controller or threshold-based automation. The agent evaluates the full environmental picture before taking any action.

Nutrient management decisions:

The agent reads pH and EC together, not independently. If pH drops and EC rises simultaneously, the most likely cause is evaporation concentrating the solution. The correct response is adding plain water to dilute back to target levels, not dosing pH-up and making the EC problem worse. A threshold controller would chase both readings separately and overdose the reservoir.

If pH drops but EC stays stable, the plants are likely consuming base cations (calcium, magnesium) from the solution. The correct response is a small pH-up dose. If EC drops while pH stays stable, the plants are consuming nutrients evenly and the solution needs a nutrient top-up.

The agent tracks these readings over time. A gradual pH drift of 0.1 over 6 hours is normal plant activity. A sudden drop of 0.5 in an hour might indicate contamination or a failed dosing pump that is leaking solution. The agent flags abnormal rates of change as alerts rather than trying to compensate automatically.

Climate management decisions:

Temperature and humidity interact through VPD (vapor pressure deficit). High VPD (hot, dry air) drives fast transpiration, which stresses plants and increases nutrient uptake rate. Low VPD (cool, humid air) slows transpiration and can promote mold. The agent calculates VPD from temperature and humidity readings and adjusts ventilation fans to keep it in the optimal 0.8-1.2 kPa range for most crops.

On a hot afternoon, the agent might open the exhaust fan to reduce temperature, but only if the outdoor humidity is lower than indoor humidity. Venting humid outdoor air into the greenhouse during a rainstorm would raise indoor humidity and drop VPD below target. The agent checks weather conditions (via an API call or an outdoor sensor) before deciding on ventilation strategy.

Lighting decisions:

The BH1750 light sensor tracks daily light integral (DLI), the total light energy received over the day. Lettuce needs around 12-17 mol/m2/day, while tomatoes need 20-30. On overcast days, the agent calculates the shortfall from natural light and schedules supplemental grow lights for the evening hours, targeting the cheapest electricity rates if time-of-use pricing applies.

Dosing safety limits: Like the irrigation controller, hard-code maximum dosing volumes in the pump control script. If the agent tries to dose 50ml of pH-down (enough to crash the solution pH), the script caps it at 5ml per dose. The agent can always dose again on the next cycle, but it cannot overdose in a single action. This is critical. Concentrated pH adjusters can kill plants in minutes if overdosed.

A greenhouse controller generates valuable data: sensor readings every few minutes, dosing events, climate adjustments, and the reasoning behind each decision. That data is useful for troubleshooting problems, optimizing growing conditions, and tracking performance across crop cycles.

Local logging: Store readings in a SQLite database on the Pi for immediate access. Each row captures a timestamp, all sensor values, any actions taken, and the agent's reasoning. SQLite handles the write volume of a greenhouse controller (a few rows per minute) without issues and does not require a separate database server.

Remote access with Fastio: A greenhouse Pi often sits in a backyard or on a rural property where you cannot easily SSH in from your phone. Fastio workspaces give you a practical way to access your greenhouse data from anywhere. The agent can upload daily summary reports, sensor CSV exports, and alert logs to a shared workspace using the Fastio MCP server.

With Intelligence Mode enabled, those uploaded files become searchable and queryable. You can ask "what was the average pH in the reservoir last week" or "show me all days where the fan ran for more than 4 hours" without downloading files and parsing them yourself. This is a genuine advantage of AI-indexed storage over a raw file share or an S3 bucket.

The Business Trial includes generous storage and monthly credits during the trial with no credit card required. A greenhouse generating daily log files and weekly summary reports will use a fraction of that over an entire growing season.

For growers running multiple greenhouses or multiple growing zones, each controller can upload to the same workspace or to separate workspaces under one org. The ownership transfer feature lets an agent create and populate the workspace, then hand access to a human grower who manages it through the web UI.

Other approaches work too. Grafana with InfluxDB gives you time-series dashboards. Home Assistant can ingest sensor data via MQTT. ThingSpeak provides a free cloud dashboard for IoT data. Fastio is simpler to set up since the MCP server handles uploads without running a separate database or dashboard server, but any persistent storage works.

Growth stage adaptation: The powerful long-term feature is adapting the agent's behavior to plant growth stages. Seedlings need lower EC (0.8-1.2 mS/cm), higher humidity, and less intense light. Vegetative growth pushes EC up (1.4-2.0) and increases light demand. Flowering and fruiting stages need the highest EC (2.0-2.5) and longest photoperiods.

Rather than hardcoding these profiles, tell the agent the current growth stage and let it adjust targets accordingly. When you transplant seedlings, update the agent's context to "seedling stage, week 1." The agent shifts its pH, EC, temperature, and light targets without any code changes. When the plants start flowering, update the context again. This is the core advantage over fixed automation: the same hardware and wiring adapts to completely different growing requirements through a context change in the agent's prompt.

Greenhouse environments are harsh on electronics. Heat, humidity, and nutrient splash create failure modes that do not exist in a typical indoor Pi project.

pH probe drift: pH probes degrade over time, especially in hydroponic solution. Plan to recalibrate monthly using pH 4.0 and 7.0 buffer solutions. The Atlas Scientific EZO board stores calibration data on the board itself, so recalibration does not require changes to your reading script. If readings become erratic even after calibration, the probe needs replacement. Most hydroponic pH probes last 12-18 months.

EC sensor fouling: Nutrient salts can deposit on conductivity probe electrodes, causing readings to drift high. Clean the probe monthly with a soft brush and dilute vinegar. If your EC readings are consistently 10-15% above what a reference meter shows, fouling is the likely cause.

Condensation on electronics: The Raspberry Pi and relay boards should live outside the growing area or in a sealed, ventilated enclosure. Conformal coating on the Pi's PCB adds another layer of protection. Sensor cables can run into the growing area, but the electronics should stay dry. A single drop of condensation on an I2C bus can cause intermittent communication failures that are maddening to debug.

Peristaltic pump calibration: Each pump model delivers a different volume per second. Calibrate by running the pump for 30 seconds into a graduated cylinder and measuring the output. Record the ml-per-second rate and hardcode it into your dosing script. Recalibrate whenever you replace tubing, because tube wear changes flow rate.

Network interruptions: If the Pi loses internet, the agent cannot make LLM API calls. Configure a fallback mode in your sensor scripts that reverts to simple threshold logic: if pH is below 5.5, dose pH-up for 2 seconds; if EC is above 3.0, add water for 10 seconds. This keeps the system running safely until connectivity returns, even if the decisions are less sophisticated.

Power failures: Use a normally-off relay configuration for dosing pumps. If power drops, pumps stop and no chemicals are added to the reservoir. When power returns, the agent should read all sensors before taking any action. Do not assume the reservoir state is the same as when the agent last checked.

Reservoir temperature control: If your reservoir consistently runs above 24C, add a small aquarium chiller or move the reservoir to a shaded location. No amount of agent intelligence can fix dissolved oxygen depletion caused by warm water. The DS18B20 sensor gives the agent visibility, but the physical solution is cooling the reservoir, not dosing more nutrients.